2nd/10_Wiki/Topics/AI_and_ML/Reconciliation.md

---
id: wiki-2026-0508-reconciliation
title: Reconciliation
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Data Reconciliation, Conflict Resolution, CRDT]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [reconciliation, crdt, distributed, finance, sync]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: typescript
  framework: yjs
---

# Reconciliation

## 매 한 줄
> **"매 두 source 의 truth 의 align"**. Reconciliation 의 financial (ledger ↔ bank statement match), data (master ↔ replica), distributed (CRDT conflict-free merge), AI (multi-agent state sync) 의 cross-domain pattern. 2026 의 local-first software 의 CRDT renaissance.

## 매 핵심

### 매 Financial reconciliation
- **3-way match**: PO ↔ receipt ↔ invoice.
- **Bank rec**: GL ↔ bank statement 의 daily.
- **Sub-ledger to GL**: AR/AP detail 의 control account match.
- **2026 tools**: BlackLine, FloQast, AI-driven (Trullion).

### 매 Data reconciliation
- **CDC (Change Data Capture)**: Debezium, source-of-truth log.
- **Lambda architecture**: batch + stream — 매 reconcile 의 daily.
- **Idempotency**: 매 retry 의 safe — UUID dedup.
- **Checksum**: row-count + hash 의 source vs target.

### 매 CRDT (Conflict-free Replicated Data Type)
- **G-Counter**: grow-only counter (sum across replicas).
- **PN-Counter**: increment + decrement.
- **OR-Set**: observed-remove set — add wins.
- **LWW-Register**: last-write-wins by timestamp.
- **RGA / Yjs Y.Text**: collaborative text.
- **Automerge**: JSON CRDT, local-first.

### 매 Modern distributed
- **Operational Transform (OT)**: Google Docs 의 legacy. Server-coord.
- **CRDT**: Notion, Linear, Figma (mixed). P2P-friendly.
- **Vector clocks**: causal ordering.
- **Hybrid Logical Clock (HLC)**: physical + logical — CockroachDB.

### 매 응용
1. Collaborative editing (Yjs, Automerge).
2. Offline-first mobile app sync.
3. Multi-region database (DynamoDB Global, Cosmos DB).
4. Bank ledger reconciliation.
5. Multi-agent LLM shared scratchpad.

## 💻 패턴

### Yjs collaborative text (TypeScript)
```typescript
import * as Y from "yjs";
import { WebsocketProvider } from "y-websocket";

const doc = new Y.Doc();
const provider = new WebsocketProvider("wss://sync.example.com",
                                        "room-1", doc);
const text = doc.getText("content");

text.observe(event => console.log("delta:", event.delta));
text.insert(0, "Hello ");
// 매 다른 client 의 concurrent insert 의 auto-merge
```

### Automerge JSON CRDT
```typescript
import * as Automerge from "@automerge/automerge";

let doc1 = Automerge.from({ tasks: [] as string[] });
let doc2 = Automerge.clone(doc1);

doc1 = Automerge.change(doc1, d => d.tasks.push("write spec"));
doc2 = Automerge.change(doc2, d => d.tasks.push("review code"));

const merged = Automerge.merge(doc1, doc2);
// merged.tasks == ["write spec", "review code"] — 매 conflict X
```

### G-Counter (CRDT)
```python
class GCounter:
    def __init__(self, node_id: str):
        self.node = node_id
        self.counts: dict[str, int] = {}
    def inc(self, n: int = 1):
        self.counts[self.node] = self.counts.get(self.node, 0) + n
    def value(self) -> int:
        return sum(self.counts.values())
    def merge(self, other: "GCounter"):
        for k, v in other.counts.items():
            self.counts[k] = max(self.counts.get(k, 0), v)
```

### Bank reconciliation pseudo
```python
def reconcile(gl: list[Txn], bank: list[Txn]) -> dict:
    matched, gl_only, bank_only = [], [], []
    bank_idx = {(t.amount, t.date): t for t in bank}
    for g in gl:
        key = (g.amount, g.date)
        if key in bank_idx:
            matched.append((g, bank_idx.pop(key)))
        else:
            gl_only.append(g)
    bank_only = list(bank_idx.values())
    return {"matched": matched, "gl_only": gl_only, "bank_only": bank_only}
```

### Idempotent upsert (Postgres)
```sql
INSERT INTO orders (id, amount, status)
VALUES ($1, $2, $3)
ON CONFLICT (id) DO UPDATE SET
  amount = EXCLUDED.amount,
  status = EXCLUDED.status,
  updated_at = NOW();
```

### CDC with Debezium (Kafka Connect)
```json
{
  "name": "pg-source",
  "config": {
    "connector.class": "io.debezium.connector.postgresql.PostgresConnector",
    "database.hostname": "db", "database.dbname": "app",
    "plugin.name": "pgoutput",
    "topic.prefix": "app",
    "table.include.list": "public.orders,public.payments"
  }
}
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| Real-time collab text | Yjs (mature, fast) |
| Offline-first app | Automerge (rich JSON) |
| Counter across regions | PN-Counter / G-Counter |
| Centralized server OK | OT (simpler) or CRDT |
| Bank/finance | Manual + AI-assist (BlackLine) |
| DB replica sync | CDC (Debezium) + idempotency |

**기본값**: Yjs 의 collab text, Automerge 의 structured offline-first, CDC 의 enterprise data sync.

## 🔗 Graph
- 부모: [[Distributed-Systems]] · [[Data-Engineering]]
- 변형: [[CRDT]]
- 응용: [[Local-First]]
- Adjacent: [[Eventual Consistency]] · [[CDC]] · [[Idempotency]]

## 🤖 LLM 활용
**언제**: 매 reconciliation rule 의 fuzzy match (description, amount tolerance), exception explanation, audit trail summary.
**언제 X**: deterministic match (use SQL JOIN), regulated final attestation (human sign-off).

## ❌ 안티패턴
- **OT 의 P2P 의 force**: 매 OT 의 server 의 require — P2P 의 CRDT 의 use.
- **LWW 의 collab text**: 매 char-level loss — RGA / Yjs 의 use.
- **No vector clock 의 causal**: 매 wall-clock skew → wrong order.
- **No idempotency key**: retry 의 duplicate → double charge.
- **Reconcile manually 매 day**: automate w/ exception queue.

## 🧪 검증 / 중복
- Verified (Yjs docs, Automerge paper, Shapiro CRDT survey 2011, Debezium docs).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — finance + CRDT + data sync unified |